Shrink the question until a thumb can answer it

TL;DR: A decision a founder keeps costs the factory the time it spends waiting for the answer, not the seconds of thought the answer takes. Cut the count of such decisions if you like, but then reshape each one that remains until it fits on a phone and a thumb can settle it.

The insight

The price of a human decision in an autonomous business is idle time, not effort. When one of the factory’s loops, the small programs that each wake on a schedule to do one job, stops to ask Peter a question, every loop queued behind it stops too, and the bill is the hours between the question leaving and the answer arriving. Whether Peter needed ten seconds or ten minutes to think barely enters into it.

Most builders see this and cut the number of decisions the human keeps. That is the right first move, and the earlier post in this series, Notes from a factory that runs itself, mostly, is about which ones to keep. This piece takes that set as given and asks a narrower question: how long does each kept decision take to come back? The answer depends less on how many there are than on the shape of each one. A question shaped for a desk waits for a desk. A question shaped for a phone waits only for a thumb.

In practice

Peter stood between the factory and most of its approvals, and each one was shaped for his laptop, because he had to open a terminal, read the context and type a reply. Code could merge on its own while Peter slept, but an approval could not, so a task that needed his yes waited until he next sat down, and the chain of loops behind it waited with it. The factory was never slow at building, only at being allowed to continue.

The fix we settled on was not a cleverer approval but a smaller one. A loop that needs Peter now sends a short question to his phone that admits only yes or no, he answers with a thumb, and the loop that asked picks up where it stopped. Nothing about the decision changed except its shape, and that alone should move the factory from waiting for a laptop to waiting for a thumb, and the next section says how we will know.

What we’ll try next

We will measure the gap. Every question that reaches the phone will record when it left the factory and when the thumb came back, so that each approval carries a wait time as well as an answer. An approval we cannot time is one we cannot shrink further, because we cannot tell which questions still stall the loops behind them and which have become cheap. Once the slow ones are visible, the next move is plain: either rephrase them until they fit on the phone, or ask whether the machine should be deciding them at all.

One honest number

Two is the number of answers the phone accepts, and that is the whole point. A thumb can settle a question with two possible answers in the time it takes to unlock a screen, whereas a question with a text box waits for a keyboard and the mood to use it. What the number does not tell us is how long each question sits on the phone before the thumb arrives, because we have not yet measured that, and I would rather admit the gap than guess at it. Until we have that figure, “faster” is a feeling, and the factory does not run on feelings.